Recursive Refinement Network for Deformable Lung Registration between Exhale and Inhale CT Scans
This work addresses the challenge of accurately aligning lung CT scans for medical imaging applications, representing an incremental improvement by revisiting and enhancing a known principle of recursive refinement.
The paper tackles the problem of deformable lung registration between exhale and inhale CT scans by proposing a recursive refinement network (RRN) that recursively refines deformation vector fields across scales, achieving a state-of-the-art average Target Registration Error (TRE) of 0.83 mm on the DirLab COPDGene dataset, which is a 13% error reduction from the best leaderboard result and an 89% reduction compared to deep-learning-based peer approaches.
Unsupervised learning-based medical image registration approaches have witnessed rapid development in recent years. We propose to revisit a commonly ignored while simple and well-established principle: recursive refinement of deformation vector fields across scales. We introduce a recursive refinement network (RRN) for unsupervised medical image registration, to extract multi-scale features, construct normalized local cost correlation volume and recursively refine volumetric deformation vector fields. RRN achieves state of the art performance for 3D registration of expiratory-inspiratory pairs of CT lung scans. On DirLab COPDGene dataset, RRN returns an average Target Registration Error (TRE) of 0.83 mm, which corresponds to a 13% error reduction from the best result presented in the leaderboard. In addition to comparison with conventional methods, RRN leads to 89% error reduction compared to deep-learning-based peer approaches.